For $f \in \mathbb{Q}[x]$, does there necessarily exist $a \in \mathbb{Z}$ such that $af \in \mathbb{Z}[x]$ and is primitive?

76 Views Asked by At

I'm working through a proof from lecture notes in an algebra class that for all $n$, the $n$-th cyclotomic polynomial $\Phi_n$ has integer coefficients. The proof proceeds as follows:

  1. Write $x^n - 1 = \Phi_n \cdot g$ with $g \in \mathbb{Q}[x]$.
  2. There are $0 \ne a, b\in \mathbb{Z}$ such that $a\cdot\Phi_n, b\cdot g \in \mathbb{Z}[x]$ and are both primitive.
  3. By Gauss's lemma, $ab(x^n - 1) = (a \Phi_n)(b g)$ is primitive, and so $ab = 1$, and as such $a = b = \pm 1$, and so $\Phi_n \in \mathbb{Z}[x]$

My issue is with step 2. I understand how you can justify this if you relax it to $a,b \in \mathbb{Q}$, but we need $a,b \in \mathbb{Z}$ for the proof. I've been looking for quite a bit, and despite this being a relatively simple lemma, I can't find a proof of it and am starting to wonder if it's even true, especially given that other proofs of the same thing seem to be much longer.

1

There are 1 best solutions below

0
On

You are right with the issue in step 2. There is a simple counterexample: $$ g(x) = 2x + \frac{2}{3} $$ However, $\Phi_n(x)$ has leading coefficient 1, and so does $g(x)$ since $x^n-1 = \Phi_n(x)\cdot g(x)$. Therefore in this particular case of the proof, $a,b\in\mathbb{Z}$ is correct.